Is an autonomous vehicle safe?
Posted on: February 18, 2019 by Fabio De Pasquale
The autonomous vehicle is seen as the solution to road traffic without accidents, but in reality, we know that this is not possible, but the elimination of human error should lead to a drastic reduction in the number of accidents.
Autonomous vehicles have shorter reaction times than ours, they respect the rules of the road and do not get angry with other drivers or pedestrians, they do not get drunk, they do not take drugs.
Their real problem is that of having to move, at least for now, in a hybrid environment, in which most of the vehicles are conducted by human beings who instead often have unpredictable and not always rational behaviors and which can therefore undermine their electronic brain.
Until fully autonomous driving (SAE level 4), factors such as road type, speed range and environmental conditions are considered when deciding whether a car can perform the task of driving.
Whose responsibility is it if an autonomous vehicle causes an accident?
There is still no clear answer to this question, because the answers could be multiple, such as the vehicle manufacturer, the provider of the artificial intelligence system that drives it, the sensory system providers or the public authorities who authorized the vehicle to circular.
The need for specific legislation in the sector
A precise regulation is therefore necessary not only to guarantee an adequate level of safety but also to make possible the definition of the related responsibilities after a possible incident.
UNECE (United Nations Economic Commission for Europe), the body that defines the legislation to be complied with by vehicle manufacturers to be approved, has acknowledged the need to amend the Vienna Convention to allow the circulation of autonomous vehicles.
Currently, the European Commission is also working to create a legislation that regulates this transport sector.
To be considered "safe" an autonomous vehicle should have a probability of being involved in a much lower accident than a vehicle driven by a person, at least, indicatively, two orders of magnitude lower.
As reported by the annual OCSE form, data in Europe are very different from country to country but generally a vehicle is involved in a fatal accident every 100-300 million km traveled, so an autonomous vehicle should not have more than one fatal accident every 10-30 billion km traveled, but currently it is possible to think of testing the vehicle for similar distances, the regulation in discussion, it is orienting itself on the following prescriptions:
- The vehicle must behave correctly in traffic, along a significant route from the point of view of the scenarios encountered (urban, suburban, etc.)
- The vehicle must avoid accidents in a series of simulated scenarios in a traffic-free lane, typical scenarios of dangerous situations that the vehicle may encounter in real traffic;
- In computer simulation, the vehicle must pass, without accidents, a long series of critical situations, detected in real traffic;
- Finally, the manufacturer must self-certify that he has developed the solution according to all the safety criteria available.
What awaits us in the future?
The new legislation should provide that the vehicle registers all the information necessary to define any liability in the event of an accident, just like the black boxes of the aircraft.
The regulatory system will then be able to "learn" from the few remaining incidents to promote a software check and thus avoid them in the future.
Second or third level autonomous driving systems, with continuous supervision by the driver, are already available or will be available in the next few years by almost all vehicle manufacturers.
For higher level systems, where the driver can devote himself completely to other activities, even sleeping for example, it will be necessary to wait a few more years: the legislation must still be well defined and the legislators will probably want to wait for the results of the extensive experiments underway in US as well as in Europe.
At the Atos Technology Days, our experts will explain how important it is to promote an ethical framework for artificial intelligence. Learn more about the event here